Qwen3.8-Max is Alibaba's newest flagship large language model, the successor to Qwen3.7-Max. It is a reasoning model — responses include an internal thinking phase, and usage reports separate reasoning_tokens from visible output tokens, just like other thinking-mode models.
In our own testing, the production API endpoint is qwen3.8-max (note: the -preview variant is restricted and returns 403 for standard API access). Responses in our checks came back in around 1.8 seconds for short prompts.
Alibaba publishes Qwen3.8-Max at the same list price as the previous generation. Pricing is unified across thinking and non-thinking output:
| Tier | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| Standard (0 – 1M tokens) | ¥12 ≈ $1.67 | ¥36 ≈ $5.00 |
| Batch API (50% off) | ¥6 ≈ $0.84 | ¥18 ≈ $2.50 |
New Alibaba Cloud accounts also receive a 1 million token free quota for Qwen3.8-Max, valid for 90 days after activation.
No — not yet. We checked OpenRouter's full model catalog on August 3, 2026: out of 337 listed models, there are zero qwen3.8 entries. The newest Qwen flagship available there is still Qwen3.7-Max:
| Model on OpenRouter | Input / 1M | Output / 1M |
|---|---|---|
| qwen/qwen3.7-max | $1.475 | $4.425 |
| qwen/qwen3.6-max-preview | $1.027 | $6.162 |
| qwen/qwen3-max | $0.78 | $3.90 |
| qwen3.8-max | Not listed | |
For context, when Qwen3.7-Max launched, it appeared on OpenRouter within about a day. Qwen3.8-Max may follow the same pattern — but right now, if you want the 3.8 generation, OpenRouter is not an option.
| Qwen3.8-Max | Qwen3.7-Max | |
|---|---|---|
| Official input / 1M | ¥12 ≈ $1.67 | ¥12 ≈ $1.67 (list) |
| Official output / 1M | ¥36 ≈ $5.00 | ¥36 ≈ $5.00 (list) |
| Current promotion | None yet | Limited-time 50% off (¥6 / ¥18) |
Same list price — but the 3.7 promotion means the older model is temporarily the cheaper choice if raw cost is your only criterion. The 3.8 generation is the pick if you want the newest reasoning capabilities at the standard rate.
Requires an Alibaba Cloud account and a DashScope API key. Billed in the official rates above; international users may face payment and invoicing friction.
NovAI serves qwen3.8-max through a direct Alibaba channel at the official $1.67 / $5.00 rate — no platform markup. One OpenAI-compatible endpoint also gives you DeepSeek, GLM, Doubao, Kimi and more under the same key:
curl https://aiapi-pro.com/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"qwen3.8-max","messages":[{"role":"user","content":"Hello"}]}'Sign-up takes an email only — new accounts get $2 free credit, and top-ups work with USDT (TRC20).
At $1.67 / $5.00, a typical agent turn with ~2,000 input tokens and ~1,000 output tokens (including reasoning) costs roughly $0.008 — about $8 per 1,000 such turns. For batch pipelines that can tolerate async processing, the Batch API halves that.
¥12 (≈$1.67) per million input tokens and ¥36 (≈$5.00) per million output tokens at the official Alibaba Cloud rate. Batch mode is half price.
Not as of August 3, 2026. We verified the live catalog: no qwen3.8 entries. Qwen3.7-Max ($1.475 / $4.425) remains the newest Qwen flagship there.
Alibaba Cloud DashScope directly, or through NovAI's OpenAI-compatible API at the official rate with USDT (TRC20) billing.
Yes — Alibaba bills thinking and reply tokens together at the output rate.